aeh

README

äh...

Just a simple front for calling out to Large Language Model (LLM) APIs, currently supporting only the OpenAI API.

Scope

Just for my personal use. I like mods for some things but I found it a touch intransparent in which model it actually uses and wanted something super-simple that basically just stripped the JSON-cruft around the actual answer in a not-completely-shabby way.

Features

• Query OpenAI's GPT models via the command line.
• Supports setting configurations via a YAML file.
• Customizable model and temperature settings.
• Records prompt and response history.
• Graceful handling of SIGINT and SIGTERM signals.

Building / Installing

It's Go, so you need go.

Quick Install

go install github.com/ja-he/aeh

This will unfortunately (or fortunately) not give you the full non-ASCII binary name experience, so you may also want to:

git clone https://github.com/ja-he/aeh äh
cd äh
go install äh.go

Manual Build

1. Ensure you have Go installed on your system.

2. Clone the repository or download the code files.

3. Navigate to the directory containing the code.

4. Run the following command to build the utility:

   go build
   

Configuration

The utility uses a configuration file in YAML format. By default, the configuration file is created in ~/.config/äh/config.yaml if it does not exist. The configuration file allows you to set default values for the model and temperature.

Example default configuration:

defaults:
  model: gpt-3.5-turbo
  temp: 0.7

Usage

To run the tool, use the following syntax:

./äh [flags] <prompt>

Flags

• -m : Specify the model to use (e.g., gpt-3.5-turbo, gpt-4).
• -t : Specify the temperature to use for the model (a value between 0.0 and 1.0).

Examples

1. Here is how you do a simple prompt:

   ./äh "A Haiku on linear programming"
   

   That will give you something like this:

   remaining requests: 199
   remaining tokens: 9975
   model: gpt-4-0613
   total tokens used: 28
   Optimizing lines,
   Boundaries clearly defined,
   Solutions we find.
   

2. As we all know, LLMs can do some really impressive things that people struggle with, and a key skill is knowing how to use them right. Accordingly, you can of course configure the temperature and model you use, for example:

   äh \
       -m 'gpt-4o' \
       -t 1.8 \
       'generate the best UUID. Just do it, no nonsense. I do not want one of those random, common UUIDs, I want a really good one. Answer with just the UUID.'
   

   remaining requests: 4999
   remaining tokens: 599945
   model: gpt-4o-2024-05-13
   total tokens used: 58
   4e495ardenl3anspw394utitkte-Uniseterminate
   

   Marvelous. 🤯

3. You can get rid of those extra lines (which are mostly informational cruft from the response headers) by simply redirecting STDERR. If something goes wrong with the request though, that might leave you blind to what it was.

   ./äh "A limerick about Langton's ant" 2> /dev/null
   

   Langton's ant, quite a curious sight
   Moves in a grid, left or right
   Black to white, white to black
   Never does it look back
   In its algorithmic, infinite flight.
   

For more, see ./äh -h.

Environment Variables

• OPENAI_API_KEY : Set this environment variable to your OpenAI API key.
• AEH_CONFIG_DIR : (Optional) Custom directory for storing configuration and history files.
• AEH_ERR : (Optional) Set to "false" to disable error messages.
• AEH_SPIN : (Optional) Set to "false" to disable the spinner animation.

Error Handling

If the OPENAI_API_KEY environment variable is not set, the utility will terminate with an error message. Ensure to set this environment variable before running the tool.

History

The tool appends each prompt and response to a history file located in the configuration directory (~/.config/äh/history.json).

License

This project is licensed under the MIT License, see LICENSE.

Contributing

Feel free to open issues or submit pull requests, but don't expect too much from me here; I am quite happy keeping this tool fairly basic (or perhaps rather, I am averse to turning this thing I just made to scratch a small itch I have from time to time into a complicated thing that diverges from my narrow expectations and needs).

Feel free to fork as well, of course, if you have a cool idea that I prefer not to have here 🙂

Contact

For any questions or feedback, please feel free to open an issue or discussion.

Why not call it 'hmm' so it could be typed on more keyboards?

äh...

Documentation

Overview

Package is a command-line utility for calling LLM APIs, currently only the OpenAI API. It is intended to be simple and straightforward first and foremost.